A database de-identification framework to enable direct queries on medical data for secondary use

Area of Science:

• Biomedical Informatics
• Data Security
• Health Data Management

Background:

• Patient clinical records require de-identification for research to minimize protected health information exposure.
• Electronic medical record (EMR) data must meet stringent privacy standards before research utilization.

Purpose of the Study:

• To demonstrate a robust framework for structured data de-identification applicable to any relational data source.
• To qualify patient clinical records as non-human-subject research data.

Main Methods:

• A pilot implementation using a real-world clinical data warehouse evaluated the de-identification process on limited subject areas.
• Compared query results and performance between source and target systems to validate data accuracy and usability.

Main Results:

• A framework combining hashing, pseudonyms, and session-dependent randomization rigorously guards against identifier exposure and re-identification.
• Includes a query rejection option to prevent identification from low-volume data, though re-identification via prior knowledge is not prevented.
• Modular design allows for future integration of natural language processing for medical free text de-identification.

Conclusions:

• A framework was demonstrated that produces HIPAA compliant databases directly queryable by researchers.
• This de-identification technique can be enhanced to support inter-institutional research data sharing via middleware like caGrid.